Abstract:
Invasive alien plants (IAPs) not only pose a serious threat to biodiversity and water resources but also have impacts on human and animal wellbeing. An important step in IAPs management is to map their location as there is a strong correlation between the spatial extent of an invaded area and the effort required for clearing the plant invasion. However, the traditional GPS based IAPs mapping field campaigns are costly, time consuming and labour intensive. The developments in the Unmanned Aerial Vehicle (UAV) technology have afforded the remote sensing (RS) community the opportunity to map IAPs at enhanced temporal and spatial resolutions. As a result, this framework synthesises a UAV-RS approach for mapping invasive alien plants in South African semi-arid woodlands using Harrisia pomanensis (the Midnight lady) as a case study. In particular, this framework outlines procedures for geometric and radiometric calibration of UAV-derived orthomosaics as well a semi-automated object-based image classification technique for mapping IAPs. The geometric calibration was conducted in the Agisoft Lens software package to determine the camera interior orientation parameters. Since sample photos of the LCD screen were taken from a short-range, there were more radial than tangential distortions. In addition, a scene illumination uniformity statistical inference allowed for the radiometric calibration of the entire scene using parameters derived from radiometric calibration targets placed only in one spot within the study area using the empirical line method (ELM). In particular, accuracy assessment of the radiometric calibration resulted in a correlation coefficient (r) value of 0.977 between in situ measured reflectance and the reflectance values derived from the calibrated image wavebands. This strong correlation validated the proposed UAV-RS ELM based radiometric calibration method for applications in semi-arid woodlands. Furthermore, out of the five evaluated image classifiers, the case study demonstrated that the object-based supervised Bhattacharya classifier which gave 90% and 95.7% producer and user accuracies, respectively, produced more accurate results for mapping Harrisia pomanensis. Even more so, an area based accuracy assessment showed that the Bhattacharya classifier mapped Harrisia pomanensis better than the Maxver classifier (i.e. the second best algorithm) with mapping accuracy averages of 86.1% and 65.2%, respectively, for all the different polygon area sizes. Future research should ascertain whethe radiometric calibration increases mapping accuracy in large scale (>100ha) UAV-RS applications.